When/Where/How
MWF 11:40-12:30 CT Fall 1997
210 Tureaud Hall
Call Number 7077 (Fall 1997)
Instructor
David M. Koppelman
What
What do you do when the world's fastest CPU takes two days to compute
the one-day weather forecast? You take ten CPU's and divide the
problem between them, this is called parallel processing. The CPU's
need to communicate while solving the problem; they do so using an
interconnection network.
The course will describe a range of interconnection designs, concentrating on what are called multistage interconnection networks. Their use in message passing and shared memory issues will be covered; treatment ranges from theoretical foundations to implementation.
Topics
Review of Parallel Processing XX Communication in shared-memory and message-passing systems. XX Parallel system models. Direct Networks: n-dimensional meshes and de-Bruijn families. MIN Basics: omega family topology and operation. Performance Analyses: analytic bounds, queuing models, simulation. Banyan and Delta Classes: topology, properties, admissible permutations. Fault-Tolerant and Multiple-Path Banyan-Like Networks Clos and Related Networks Generalized Connectors Sorting and Counting Networks. Congestion Reduction Techniques Shared Memory Topics (Time Permitting)
Texts
(Recommended, about 15% overlap with this course.)
F. Thomas Leighton, "Introduction to parallel algorithms and architectures:
arrays, trees, hypercubes."
Journal and Conference Papers
Grading
30% Midterm Exam; 30% Final Exam; 20% Project; 20% Homework
Late-homework and project penalty: 10% per day late deducted. Missed-midterm-exam policy: at instructor's discretion either a makeup exam, use final exam grade for midterm grade (i.e., 60% final exam weight), or use of zero for midterm grade. Attendance: optional, however students are responsible for all material, instructions, and notices presented in class.
David M. Koppelman - koppel@ee.lsu.edu | Modified 11 Jul 1997 13:52 (18:52 UTC) |